Heretofore, in systems including multiple apparatuses, apparatuses are monitored so that a significant failure does not occur. In addition, it is known that many system failures are caused by setting errors in a system configuration.
For example, when a system implemented by using different middleware products provided by a wide variety of different vendors performs a task involving the different middleware products, integrity of setting values of parameters is to be maintained among the middleware products. Accordingly, the following technique has been disclosed: one sentence is read from a sentence table in which multiple sentences, each of which describes a relationship between two parameters, are recorded; a pattern table including multiple patterns obtained through classification of sentences is referred to; and the sentence that has been read is compared with each pattern that is obtained through classification and that is stored in the pattern table. In the technique, a pattern for which similarity to the sentence that has been read is highest is determined. Among the records in a summary table in which relationships between two parameters are summarized, in a record corresponding to the pattern for which the similarity is highest, the count value for the relationship between two parameters which is indicated by the pattern is incremented. Thus, an event having a high probability of resulting in occurrence of a failure is inferred.
Widespread use of technique such as cloud causes multiple techniques to be combined for use with one another in a layered manner. Therefore, a combination of multiple factors may cause a failure in a system such that, when such a failure occurs, it becomes difficult to determine the reason for the failure.
To address this problem, there has been used a method in which information about failures that have occurred in the past and information about actions taken in response to the failures are recorded and analyzed.
For example, the following technique has been proposed. When a range in operation history in which classification of operation history is to be performed is given, pieces of operation history which correspond to the range are obtained from an operation history database (DB) storing operation history for operation target apparatuses which has been written in the past. In this technique, feature values are extracted from each piece of operation history, and the pieces of operation history are classified, for each of units causing failures, based on similarity of the feature values.
Japanese Laid-open Patent Publication No. 2015-232757 and Japanese Laid-open Patent Publication No. 2015-153078 are examples of the related art.